Today, there is an increased need to be able to capture specific information from documents. For example, a company may want to capture a total invoice value from a group of invoices in order to automate the payment process. Existing systems are able to capture specific information from a document based on defining a specific location in the document to look for the information. For example, a value of a check can be determined by capturing a specific field from a check, such as the amount field. These types of systems work as long as the location of the field is always at the same position in the document. However, this can be problematic when a document becomes misaligned, scanned at a different resolution, or scanned upside down. In addition, in some documents, the area associated with the field may not be in the same location or even the same page due to variations in the documents, such as long lists of purchased items in an invoice. Since the specific information is no longer at the same location, existing solutions capture invalid information in these situations. What is needed is a better way to identify specific information in documents to improve overall accuracy.